Tags: Language technology

Sarvam AI, a Bengaluru‑based startup, says its Sarvam Vision model outperforms global rivals Gemini and ChatGPT on key optical character recognition (OCR) benchmarks for Indian languages. The model supports all 22 scheduled Indian languages and can handle complex tables, charts, and real‑world scene text. Paired with the Bulbul V3 text‑to‑speech system, which offers 35 local‑accented voices, the company positions itself as a builder of "sovereign AI" tailored to India’s linguistic diversity. Sarvam hopes its technology will help small businesses and government agencies digitize records more accurately and spur broader AI innovation focused on regional needs. Lire la suite

Enterprise AI firm Cohere launched the Tiny Aya family of open-weight multilingual models, supporting over 70 languages and designed for on‑device use. The base model contains 3.35 billion parameters and runs on everyday hardware without internet connectivity. Regional variants target African, South Asian, and Asia‑Pacific/West‑Asia/European languages. Trained on a single cluster of 64 H100 GPUs, the models are available on HuggingFace, the Cohere platform, Kaggle and Ollama, with accompanying datasets and a forthcoming technical report. Cohere also highlighted strong financial performance and a pending public‑market plan. Lire la suite

OpenAI has introduced a new translation interface at chatgpt.com/translate, offering users a simple text‑to‑text translation tool that mirrors familiar services like Google Translate. The page supports translation in 50 languages and links directly to ChatGPT's main chat interface, where users can refine translations, adjust tone, or request explanations suitable for children. Sample prompts appear as one‑click buttons, and the system also hints at future capabilities such as voice and image inputs. The rollout arrives as Google expands its own AI‑driven translation features, highlighting a competitive push in the language‑technology market. Lire la suite

A recent hands‑on test of Apple’s new live‑translation feature, built into the latest AirPods Pro 3, showed how the technology can break language barriers during a family visit. Paired with an iPhone running the latest iOS, the system provided real‑time subtitles for a Spanish‑speaking mother‑in‑law, though occasional mistranslations highlighted its beta status. The experience demonstrates both the promise of seamless, screen‑free translation and the current need for refinement. Lire la suite

Timekettle has unveiled the W4 AI Interpreter Earbuds, a new pair of real‑time translation earbuds that use bone‑enabled microphones and AI‑driven Babel OS 2.0 software. The earbuds translate speech across 42 languages and 95 accents with up to 98 percent accuracy, offering up to four hours of continuous translation and a charging case that extends use to ten hours. Priced at $349, the W4 AI provides a more casual design than the earlier W4 Pro, while still allowing custom lexicons and standard music playback for up to eight hours. Lire la suite